feat(review): place the boundary-trust, state-scope, no-lock and converge-on-rerun principle parts in the reviewer, coupling and loop-lane #12147

Parent #7589 decides that pstack's principles are placed "as reasoned sentences in the phase-skill step that applies them" (D58, Q174, Q175). The owner approved this 2026-10-15 (poteto-review decision rows 58 and 105). The wording rule the owner accepted (Q175) is that each placed part states the reason and the test that shows it applies, rather than a bare rule, and names the standard or discipline it comes from.

This slice places these parts:

• 2b and 2g (boundary-discipline), in review:code-reviewer: inside a validated boundary, trust the types, with no re-validation and no deep nil checks; the test question is "Is this data crossing a system boundary right now?" The code-reviewer agent (plugins/review/agents/code-reviewer.md:73-80) has the pure-function question at :77 but not the boundary one. The implement side landed at plugins/implementation/skills/implement/SKILL.md:52.
• 14e (minimize-reader-load), in the reviewer's mutable-state item: shrink state scope, preferring returns over mutation, locals over fields, and fields over globals. Architecture's scan has it at plugins/architecture/skills/improve/research/deepening/scan-briefing.md:41; the reviewer does not.
• 20e (separate-before-serializing-shared-state), quoted as written in /coupling:reduce rung 1 (plugins/coupling/skills/reduce/reference/remediations.md:38-47): "Instructions and conventions are not concurrency control."
• 12e (make-operations-idempotent), in the loop-lane contract: every step that writes to the forge converges when rerun. docs/contracts/loop-lane/README.md covers restart adoption (:949) and duplicate-escalation suppression (:518), but has no general rule.

Done when

• plugins/review/agents/code-reviewer.md's code-quality list gains one item: re-validation or deep nil checks on data that crossed no system boundary, with the question "Is this data crossing a system boundary right now?" and the reason (checks inside a typed boundary add noise and false safety).
• The reviewer's mutable-state item names the scope order (returns over mutation, locals over fields over globals), with the reason that a reader must track every place that can change the value.
• plugins/coupling/skills/reduce/reference/remediations.md rung 1 carries the sentence "Instructions and conventions are not concurrency control." with one line of reason. When that file has plugin copies listed in scripts/shared-copies.txt, the copies match.
• docs/contracts/loop-lane/README.md states that every step that writes to the forge (a label, comment, branch or PR) must leave the same state when rerun after a stop at any point, and that it reads current forge state before writing. It gives the reason: lanes restart and are re-dispatched. When that file has plugin copies listed in scripts/shared-copies.txt, the copies match.
• .changes/review/*.md and .changes/coupling/*.md fragments with bump: patch describe the additions, plus one for each plugin whose loop-lane copy changed.
• The PR's CI is green.

Out of scope

• Gated part 24i (optional-field bags that allow contradictory states): gated per refactor(pstack): cut marginal sections from four skills cursor/plugins#634 (Q557).
• Changing loop-lane behavior or code: this adds a design rule to the contract text only.

Parent: #7589